Divya van Mahajan

How to Make ChatGPT Ads Not Suck

ai-daily-brief-podcast

Overview

This talk, presented on The AI Daily Brief (a daily podcast and video covering AI news), responds to OpenAI’s January 2026 announcement that advertising is coming to ChatGPT’s free and “go” tiers. The central thesis is that while advertising in ChatGPT is commercially inevitable, OpenAI has a genuine opportunity—and obligation—to build an ad system that is fundamentally more user-centric, transparent, and value-additive than anything that has come before. The speaker argues that simply making ads “unobtrusive and clearly labeled” is not enough, and proposes a concrete strategic framework for how ChatGPT ads could avoid the pitfalls of Google and Meta’s advertising trajectories.

Source video URL not provided.

Prerequisites

  • Basic familiarity with how digital advertising works (display ads, cost-per-click, cost-per-impression, affiliate marketing)
  • Awareness of OpenAI’s ChatGPT product and its free vs. paid tier structure
  • General understanding of how Google Search and Meta/Facebook advertising systems function
  • Familiarity with concepts such as ARPU (Average Revenue Per User), intent-based advertising, and programmatic ad models
  • Background knowledge of the competitive AI landscape (Gemini, etc.)

Main Points

OpenAI’s Announcement and Stated Principles

  • OpenAI announced it would begin testing ads in ChatGPT’s free and “go” tiers in “coming weeks.”
  • Key stated principles: responses will not be influenced by ads; ads are always clearly labeled and separate; advertiser conversations are private; paid tiers (Plus, Pro, Business, Enterprise) will be ad-free.
  • The company frames advertising as a mission-aligned mechanism to ensure broad, equitable access to high-quality AI—not as a revenue-maximization play.
  • Sam Altman and CEO of Applications Fiji Simo both publicly emphasized that ads will not influence ChatGPT’s answers.
  • OpenAI hinted at more creative, conversational ad formats in the future (e.g., users asking questions directly within an ad unit before making a purchase).

The Commercial Inevitability

  • Only approximately 5% of ChatGPT users convert to paid tiers, making an ad-supported model financially necessary despite annualized revenue reaching ~$20 billion.
  • An anonymous analyst calculated that if OpenAI reaches one billion free users and captures just 9% of Meta’s ~$58 per-user ARPU, that equals ~$5 billion in incremental annual ad revenue; full parity would be ~$57 billion.
  • ChatGPT traffic converts at approximately 16% versus Google Organic’s ~1.76%—a roughly 9x higher intent signal—making the ad inventory potentially far more valuable per impression.
  • Ben Thompson (Stratechery) argued OpenAI should have launched ads in 2023; by now the product would be mature and the backlash absorbed.

The Trust and Credibility Problem

  • Sam Altman previously called “ads plus AI” uniquely unsettling and described advertising as a “last resort” (May 2024); his reversal has damaged credibility.
  • OpenAI reportedly denied or dismissed reports of ad testing while simultaneously developing the capability, fueling accusations of dishonesty.
  • Critics draw a direct historical parallel to Google Search: ads began as clearly separated units and gradually became nearly indistinguishable from organic results. The concern is that ChatGPT follows the same “boiling the frog” trajectory.
  • Additional concerns include: memory features increasing manipulation risk, the chilling effect on recruiting top talent, and a general erosion of OpenAI’s positioning as an “ethical AI alternative.”
  • One noted upside: advertiser pressure historically exerts a normalizing effect on platform behavior, requiring minimum standards of conduct.

User Control as a Foundation

  • The speaker argues stated principles are insufficient without granular, legible user controls:
    • Users should be able to see and correct the system’s model of their preferences.
    • Users should be able to interact with ads as partners (e.g., “I already bought that” or “I’m researching, not buying—check back in three weeks”).
    • A flag-and-skip mechanism should be implemented: flagging an insensitive or mistimed ad earns the user a day of ad-free experience, putting OpenAI’s money where its mouth is.
    • Transparent advertiser ratings: user ratings of ads should carry financial consequences—lower-rated ads cost more per impression; higher-rated ads get preferential access and lower rates. Aggregate quality scores should be made public.

Innovative Ad Unit Formats

The speaker proposes five categories of ad innovation:

1. Transactional Advertising (Pay for Results)

  • Shift from cost-per-click/view (pay-for-attention) to pay-for-verified-outcomes (e.g., $50 per completed booking vs. $2 per click).
  • Introduces a user-initiated “help-me-buy” mode—a deliberate context switch that makes the user an active participant rather than an interruption target.
  • Buying agents that negotiate and filter options on the user’s behalf, with transparent disclosure of when and how advertisers pay.

2. Offers Exchange

  • Contextual, memory-triggered offers served in-conversation (e.g., “Six weeks ago you said you’d wait for a Nike sale—it’s 30% off today”).
  • A browsable offers exchange—a dedicated space to track and explore current deals, analogous to coupon or deal aggregator sites.
  • Offers include verified scarcity data, price history context, and negotiation features.
  • Users have immediate controls to adjust offer preferences inline.

3. Brand Advertising That Funds Capabilities

  • Brands fund expanded access rather than buying click-through placements—e.g., McKinsey sponsors additional deep research queries for free users who have hit their limit.
  • “Training mode presented by Nike”—a branded experience triggered by relevant athletic goal conversations.
  • Brands can fund features that would not otherwise exist for free users, creating genuine value exchange.

4. Branded Action Agents

  • The ad unit itself becomes a product: brands build constrained mini-apps inside ChatGPT that users explicitly opt into for a discrete purpose.
  • Examples: American Express Travel Concierge, TurboTax Tax Prep Assistant.
  • Users enter the branded environment, use it, and exit; the UI can expand to encompass the brand experience.
  • Identified as the highest near-term opportunity because it can immediately demonstrate utility to users.

5. AI Founders and Small Business Grants

  • A grants program providing ad credits to small businesses and AI-native startups.
  • Creates a “Kickstarter energy”—ads become stories about a new generation of businesses rather than interruptions.
  • Grant recipients are publicly browsable by category (AI-built startups, local businesses, creator businesses), giving users a reason to engage with the ad ecosystem rather than resist it.

Key Concepts

  • ARPU (Average Revenue Per User): The average advertising revenue generated per user per year; Meta’s was ~$58 in 2025.
  • Intent signal: A measure of how likely a user is to take a purchasing action based on their current activity; ChatGPT users exhibit significantly higher intent than typical search users.
  • Pay-for-verified-outcomes: An ad pricing model where advertisers pay only upon confirmed transaction completion rather than for impressions or clicks.
  • User-initiated commercial mode (“help-me-buy”): A deliberate, user-triggered context switch in which the AI actively assists with purchasing decisions, reframing ads as a feature rather than an interruption.
  • Offers exchange: A browsable, dedicated interface within ChatGPT where users can proactively explore current advertiser offers, triggered by memory and context.
  • Brand capability funding: A format where brands sponsor expanded AI capabilities (e.g., more deep research queries) for free users rather than purchasing traditional display placements.
  • Branded action agents: Mini-apps built by brands inside ChatGPT that users explicitly opt into for a specific, bounded purpose; the ad unit functions as a product.
  • Flag-and-skip mechanism: A user control that allows flagging of poor or mistimed ads, with a tangible consequence (ad-free period) as accountability.
  • Transparent advertiser ratings: A public quality-scoring system for ads where user satisfaction ratings directly affect advertiser pricing and access.
  • Boiling the frog: A metaphor for the gradual, incremental integration of ads into core product experiences, such that users normalize each incremental step; cited as Google Search’s historical trajectory.

Summary

The speaker accepts that advertising in ChatGPT is commercially inevitable given OpenAI’s burn rate and the low conversion rate of free users to paid tiers, but argues that the company has both the opportunity and the obligation to build a fundamentally better ad system rather than replicating the Google or Meta playbook. Credibility has already been damaged by inconsistent public messaging, and the greatest risk is that ChatGPT follows the historical pattern of gradually integrating ads deeper into responses until they are indistinguishable from organic answers. To avoid this, the speaker proposes a two-layer solution: first, a genuine user control foundation—including preference transparency, correction mechanisms, a flag-and-skip policy with real consequences, and public advertiser quality ratings—that goes beyond stated principles; and second, a portfolio of innovative ad formats including pay-for-verified-outcomes transactional ads, memory-triggered offers exchanges, brand-funded capability expansions, opt-in branded action agents, and a grants program for AI-native small businesses. The central argument is that OpenAI should set an ambitious goal not merely to make ads tolerable, but to make them genuinely value-additive—leveraging ChatGPT’s unique intent signal and conversational context to create an ad experience unlike anything that has existed before.